Eliminating invariances by preprocessing for kernel-based methods

1. An improved method for implementing a local invariance for use in a kernel-based classifier system, the improvement comprising the step of: incorporating the local invariance in such a way that a resulting dimension of each feature vector is fixed and wherein that dimension is equal to the dimension of input data minus the number of degrees of freedom in the local invariance; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

2. Apparatus for use in a kernel-based classifier system, the apparatus comprising: a preprocessor, which operates on input data in such a way as to provide any local invariance in providing data to the kernel-based classifier system; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

3. The apparatus of claim 2 wherein the kernel-based classifier system is a support vector machine.

4. The apparatus of claim 2 wherein the provided data has an additive invariance property such that the kernel-based classifier system only has to learn a vector-based function that is independent of values of the input data when the input data is translated.

5. A method of preprocessing for use in a kernel-based classifier system, the method comprising the steps of: receiving input data; and operating on the received input data in such a way as to provide a local invariance, other than translation, scale and rotation invariance, in providing data for use by the kernel-based classifier system; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

6. The method of claim 5 wherein the kernel-based classifier system is a support vector machine.

7. The method of claim 5 wherein the provided data has an additive invariance property such that the kernel-based classifier system only has to learn a vector-based function that is independent of values of the input data when the input data is translated.

8. A method of preprocessing for use in a kernel-based classifier system, the method comprising the steps of: receiving input data; and operating on the received input data in such a way as to provide a local invariance, besides translation, scale and/or rotation invariance, in providing data to the kernel-based classifier system; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

9. The method of claim 8 wherein the kernel-based classifier system is a support vector machine.

10. A method of preprocessing for use in a kernel-based classifier system, the method comprising the steps of: receiving binned continuous input data; and operating on the received binned continuous input data in such a way as to provide at least one local invariance in providing data to the kernel-based classifier system; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

11. The method of claim 10 wherein the kernel-based classifier system is a support vector machine.

12. A method of preprocessing for use in a kernel-based classifier system, the method comprising the steps of: receiving input data; and operating on the received input data in such a way as to provide a number of local invariances such that at least one of the local invariances is not translation, scale and/or rotation invariance, in providing data to the kernel-based classifier system; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.

13. The method of claim 12 wherein the kernel-based classifier system is a support vector machine.

14. Apparatus for use in a kernel-based classifier system, the apparatus comprising: a preprocessor, which operates on input data in such a way as to provide a set of local invariances where the set of local invariances includes a local invariance which is not translation, scale, or rotation invariance; wherein the input data is of dimension N and the provided data is of dimension M, where M<N.